三維模型的局部匹配和檢索方法研究
發(fā)布時間：2018-12-14 03:41
【摘要】：隨著三維模型技術(shù)和互聯(lián)網(wǎng)技術(shù)的不斷發(fā)展，越來越多的三維模型制作軟件和三維模型文件被共享在互聯(lián)網(wǎng)上面，同時三維模型技術(shù)的應(yīng)用領(lǐng)域也越來越廣泛，如產(chǎn)品設(shè)計、制造仿真、虛擬現(xiàn)實、3D網(wǎng)絡(luò)游戲等。尤其是最近幾年3D打印機(jī)的出現(xiàn)，三維模型的應(yīng)用已經(jīng)開始普及至家庭用戶，使得家庭用戶可以應(yīng)用3D打印機(jī)打印自己所需的三維模型。因此，研究和開發(fā)三維模型搜索引擎幫助企業(yè)用戶、家庭用戶快速、準(zhǔn)確的檢索到自己所需的三維模型，是最近幾年的研究熱點之一。 論文針對三維模型特征描述和檢索這一問題展開研究，主要工作包括以下幾個方面： 研究了基于形狀統(tǒng)計方法的特征提取算法，針對傳統(tǒng)的D2形狀描述算法的不足，提出了一種基于面積算子的三維模型檢索算法，該算法首先將對三角網(wǎng)格模型進(jìn)行模型簡化處理，使得表達(dá)三維網(wǎng)格模型的點和面的集合達(dá)到最小化，然后對三維網(wǎng)格模型上的頂點進(jìn)行統(tǒng)計分析，對每個頂點所關(guān)聯(lián)的三角形的面積進(jìn)行計算，并對點相關(guān)的面積進(jìn)行歸一化處理，然后對點關(guān)聯(lián)的面積序列進(jìn)行通傅里葉變換，得到特征向量，做出面積分布圖，通過計算模型間特征向量的差異，得到模型間的差異，進(jìn)而檢索出相似的三維模型。實驗表明，和類似的基于形狀統(tǒng)計的模型檢索算法相比較，該方法可以獲得更好的檢索結(jié)果。 針對三維模型檢索中的局部特征描述和匹配問題，提出了一種基于頂點鄰域?qū)傩缘娜S模型檢索算法，該算法首先統(tǒng)計三維網(wǎng)格模型的相關(guān)屬性和頂點的鄰域頂點相關(guān)屬性，包括鄰域的質(zhì)心、頂點到鄰域質(zhì)心的矢量、頂點到鄰域質(zhì)心的距離、頂點的法向矢量、頂點的曲率以及頂點的法向矢量和頂點鄰域質(zhì)心矢量之間的夾角等，接著對鄰域頂點的屬性中的2個角度進(jìn)行16等分，形成一個16×16規(guī)模的特征矩陣，然后采用一組特定的矩陣相似度計算方法計算特征矩陣間的相似度，，最后計算模型整體的相似度，以此來代替二個三維模型的相似度。通過和其他檢索算法的比較實驗表明，該方法對具有較豐富局部特征的模型可以獲得更好的檢索結(jié)果。 提出了一種基于法向夾角直方圖的三維模型檢索算法。算法首先對三維模型進(jìn)行預(yù)處理，接著定義了三維模型上的三角網(wǎng)格各個頂點的法向與三角網(wǎng)格之間的夾角計算方法，然后根據(jù)三角網(wǎng)格三個頂點的法向與三角網(wǎng)格之間的夾角，對三角網(wǎng)格進(jìn)行分類，根據(jù)夾角是銳角還是鈍角，將三角網(wǎng)格分成四種類型，對每種類型的三角網(wǎng)格集合構(gòu)造形狀分布曲線，通過對模型的四條形狀分布曲線的比較，得出兩個模型的相似度，從而實現(xiàn)模型的相似性檢索。實驗表明，該算法的檢索準(zhǔn)確率和檢索效率方面要由于其他類似的直方圖算法。 在三維模型的局部匹配和檢索方面，引入了三維分割技術(shù)，首先采用基于Laplace-Beltrami算子特征函數(shù)的模型特征描述方法，采用有限元算子方法離散化和求解Laplace-Beltrami算子特征值問題，將算子特征函數(shù)值作為三維模型的特征描述符；然后采用K均值聚類方法，對特征描述符進(jìn)行聚類分析，聚類后的結(jié)果將三維模型分割成多個局部區(qū)域；最后采用指派問題中的匈牙利算法計算三維網(wǎng)格模型之間的局部區(qū)域與整體模型之間的匹配度，從而得到模型間的匹配結(jié)果。 為了驗證前面提到的算法，設(shè)計了一個三維模型檢索系統(tǒng)的原型系統(tǒng)，系統(tǒng)中提供了多種檢索算法，包括面積分布算子算法、頂點鄰域?qū)傩运惴ê突贚B算子特征函數(shù)的算法以及課題組其它成員設(shè)計的算法。
[Abstract]:With the development of the three-dimensional model technology and the Internet technology, more and more three-dimensional model production software and three-dimensional model files are shared on the Internet, and the application field of the three-dimensional model technology is also more and more widely, such as product design, manufacturing simulation, virtual reality, and the like. In particular, the appearance of the 3D printer in recent years, the application of the three-dimensional model has started to spread to the home user, so that the home user can use the 3D printer to print the three-dimensional model required by the user. Therefore, the research and development of the three-dimensional model search engine to help the enterprise users, home users to search quickly and accurately to the three-dimensional model that they need, is one of the research hot spots in recent years. This paper studies the characteristics of the three-dimensional model and the retrieval of this problem. The main work includes the following: In this paper, the feature extraction algorithm based on the shape statistic method is studied, and the traditional D2 shape description algorithm is used. In this paper, a three-dimensional model retrieval algorithm based on the area operator is proposed. The algorithm first will model the triangular mesh model, so that the set of points and faces expressing the three-dimensional mesh model is minimized, and then the vertexes on the three-dimensional mesh model are unified. The method comprises the following steps of: calculating an area of a triangle which is associated with each vertex, carrying out normalization processing on the area related to a point, performing a Fourier transform on the area sequence associated with the point, obtaining a feature vector, the difference between the models is obtained to obtain the difference between the models, and then the similar three are retrieved. The experimental results show that the method can be better tested compared with similar model retrieval algorithms based on shape statistics. As a result, a three-dimensional model retrieval algorithm based on the vertex-neighborhood attribute is proposed for the local feature description and the matching problem in the three-dimensional model retrieval. The algorithm first counts the related attributes of the three-dimensional mesh model and the adjacent vertex-related attributes of the vertex. the centroid of the neighborhood, the vector of the vertex to the centroid of the neighborhood, the distance from the vertex to the centroid of the neighborhood, the method of the vertex, the curvature of the vertex and the angle between the method of the vertex and the centroid vector of the vertex, and then the two angles in the attribute of the adjacent vertex Line 16, etc., form a 16-16-16-scale feature matrix, and then calculate the similarity between the feature matrices using a set of specific matrix similarity calculation methods, and finally calculate the similarity of the whole model to replace the two three-dimensional models. The similarity of the model can be obtained by comparison with other search algorithms. The three-dimensional angle histogram based on the method is presented in this paper. The method comprises the following steps of: pre-processing a three-dimensional model, and then defining an included angle calculation method between the method of each vertex of the triangular mesh on the three-dimensional model and the triangular mesh, The angular mesh is classified, and the triangular mesh is divided into four types according to the acute angle or the obtuse angle, and the shape distribution curve is constructed for each type of triangular mesh set. By comparison of the four shape distribution curves of the model, the similarity of the two models is obtained, and the model is realized. The result of the experiment shows that the retrieval accuracy and the retrieval efficiency of the algorithm are due to other classes In terms of the local matching and retrieval of the three-dimensional model, the three-dimensional segmentation technique is introduced. First, the model feature description method based on the characteristic function of the Laplace-Beltrami operator is adopted. The method of finite element operator is used to discretize and solve the Laplace-Beltra. the characteristic function value of the operator is used as the characteristic descriptor of the three-dimensional model, the K-means clustering method is adopted, the characteristic descriptor is subjected to cluster analysis, and the result of the clustering is the three-dimensional model and finally, calculating the matching degree between the local area and the whole model between the three-dimensional mesh model by using the Hungarian algorithm in the assignment problem, In order to verify the above-mentioned algorithm, a prototype system of a three-dimensional model retrieval system is designed, and a variety of search algorithms are provided in the system including the area distribution operator algorithm, the vertex neighborhood attribute algorithm and the algorithm based on the characteristic function of the LB operator and the class.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2014
【分類號】：TP391.41
[Abstract]:With the development of the three-dimensional model technology and the Internet technology, more and more three-dimensional model production software and three-dimensional model files are shared on the Internet, and the application field of the three-dimensional model technology is also more and more widely, such as product design, manufacturing simulation, virtual reality, and the like. In particular, the appearance of the 3D printer in recent years, the application of the three-dimensional model has started to spread to the home user, so that the home user can use the 3D printer to print the three-dimensional model required by the user. Therefore, the research and development of the three-dimensional model search engine to help the enterprise users, home users to search quickly and accurately to the three-dimensional model that they need, is one of the research hot spots in recent years. This paper studies the characteristics of the three-dimensional model and the retrieval of this problem. The main work includes the following: In this paper, the feature extraction algorithm based on the shape statistic method is studied, and the traditional D2 shape description algorithm is used. In this paper, a three-dimensional model retrieval algorithm based on the area operator is proposed. The algorithm first will model the triangular mesh model, so that the set of points and faces expressing the three-dimensional mesh model is minimized, and then the vertexes on the three-dimensional mesh model are unified. The method comprises the following steps of: calculating an area of a triangle which is associated with each vertex, carrying out normalization processing on the area related to a point, performing a Fourier transform on the area sequence associated with the point, obtaining a feature vector, the difference between the models is obtained to obtain the difference between the models, and then the similar three are retrieved. The experimental results show that the method can be better tested compared with similar model retrieval algorithms based on shape statistics. As a result, a three-dimensional model retrieval algorithm based on the vertex-neighborhood attribute is proposed for the local feature description and the matching problem in the three-dimensional model retrieval. The algorithm first counts the related attributes of the three-dimensional mesh model and the adjacent vertex-related attributes of the vertex. the centroid of the neighborhood, the vector of the vertex to the centroid of the neighborhood, the distance from the vertex to the centroid of the neighborhood, the method of the vertex, the curvature of the vertex and the angle between the method of the vertex and the centroid vector of the vertex, and then the two angles in the attribute of the adjacent vertex Line 16, etc., form a 16-16-16-scale feature matrix, and then calculate the similarity between the feature matrices using a set of specific matrix similarity calculation methods, and finally calculate the similarity of the whole model to replace the two three-dimensional models. The similarity of the model can be obtained by comparison with other search algorithms. The three-dimensional angle histogram based on the method is presented in this paper. The method comprises the following steps of: pre-processing a three-dimensional model, and then defining an included angle calculation method between the method of each vertex of the triangular mesh on the three-dimensional model and the triangular mesh, The angular mesh is classified, and the triangular mesh is divided into four types according to the acute angle or the obtuse angle, and the shape distribution curve is constructed for each type of triangular mesh set. By comparison of the four shape distribution curves of the model, the similarity of the two models is obtained, and the model is realized. The result of the experiment shows that the retrieval accuracy and the retrieval efficiency of the algorithm are due to other classes In terms of the local matching and retrieval of the three-dimensional model, the three-dimensional segmentation technique is introduced. First, the model feature description method based on the characteristic function of the Laplace-Beltrami operator is adopted. The method of finite element operator is used to discretize and solve the Laplace-Beltra. the characteristic function value of the operator is used as the characteristic descriptor of the three-dimensional model, the K-means clustering method is adopted, the characteristic descriptor is subjected to cluster analysis, and the result of the clustering is the three-dimensional model and finally, calculating the matching degree between the local area and the whole model between the three-dimensional mesh model by using the Hungarian algorithm in the assignment problem, In order to verify the above-mentioned algorithm, a prototype system of a three-dimensional model retrieval system is designed, and a variety of search algorithms are provided in the system including the area distribution operator algorithm, the vertex neighborhood attribute algorithm and the algorithm based on the characteristic function of the LB operator and the class.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2014
【分類號】：TP391.41
【參考文獻(xiàn)】
相關(guān)期刊論文 前10條
1 張玉鵬;周明全;李暉;;基于形狀相似的三維模型檢索方法研究[J];北京師范大學(xué)學(xué)報(自然科學(xué)版);2011年04期
2 胡敏;羅s
本文編號：2377850
本文鏈接：http://sikaile.net/kejilunwen/sousuoyinqinglunwen/2377850.html
最近更新
教材專著
